PhD Defense by Kristoffer Stensbo-Smidt: Recycling on a Cosmic Scale

On 6 January 2017, Kristoffer Stensbo-Smidt will defend his PhD Thesis.


Recycling on a Cosmic Scale: Extracting New Information from Old Data Sets


Astronomy is at this very moment undergoing a paradigm shift. Transitioning from a time of limited data to a time of data so plentiful that it takes dedicated efforts just to store and access it. As a consequence of this, the field of astroinformatics or astrostatistics is evolving – an interdisciplinary field of astronomers, statisticians, computer scientists and data scientists.

Advanced methods from machine learning and computer vision have slowly entered astronomical research in the past couple of decades, but there is still much to do. Many open questions in astronomy could benefit from the advanced statistical methods available in the computer science field, and many interesting problems in astronomy could spark new ideas and approaches in both computer science and statistics communities.

A core hypothesis of this thesis is the idea that there is much more to be learnt from already available data sets, and that advanced statistical methods, such as machine learning and computer vision, can help uncover this information. We investigate this in three different projects covering texture analysis of galaxies, feature selection for redshift estimation, and quality assessment of images of quasar candidates.

The visual appearance of a galaxy is referred to as its morphology. A new texture descriptor for parametrising galaxy morphology is presented. The descriptor is shown to extract information about a galaxy’s specific star formation rate purely from images - information that the usual spectral energy distribution (SED) fitting misses.

Selecting the right features, for example, colours or magnitudes, for a particular task can be difficult and often relies on which have traditionally been used. An entirely general method for feature selection is introduced and shown to increase the accuracy of both redshift and specific star formation estimations.

When searching the sky for quasars, detection pipelines produce thousands of candidates, many of which must be manually inspected and evaluated. An image analysis pipeline for automatically assessing the quality of a candidate is presented and shown to be able to detect the most common situations of false positive quasar candidates.

Assessment Committee

  • Chairman: Associate Professor Jon Sporring, Department of Computer Science, University of Copenhagen
  • Senior research fellow Manda Banerji, University of Cambridge, UK
  • Professor Giuseppe Longo, University of Napoli Federico II, Italy

Academic supervisor

  • Associate Professor Kim Steenstrup Pedersen, Department of Computer Science, University of Copenhagen

For an electronic copy of the thesis, please contact